Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Date: Wed, 6 Feb 2013 22:20:35 +0100
From: Nico Golde <oss-security+ml@...lde.de>
To: oss-security@...ts.openwall.com
Subject: CVE id request: openssh?

Hello,
years ago CVE-2006-1206 was raised for a denial of service attack against 
dropbear based on exhausting the maximum number of connections.
Back in 2010 I played around with this in openssh to find out if similar 
attacks work against that. Since then I never really knew what to do with 
this, but every now and then I remember it and after this bugged me for a 
while, I finally brought up the topic to the openssh developers.

The attached program demonstrates a similar attack against a default openssh 
installation. The program simply connects to an ssh server and waits for the 
socket to be closed, thus determining the LoginGraceTime setting of the 
server. Next, it opens up connections to the server, keeping them open until 
no further connection is allowed and thus determining the MaxStartUps setting 
(of course, this may not be always accurate depending on the currently active 
sessions etc, but this is a minor detail).

The code continues to sleep for logingracetime seconds and spawns maxstartup 
connections again. As a result, unless you are very lucky and you hit the time 
window between the connection respawn, a user can not login anymore.

While this is a standard problem for any network service that limits the 
number of connections, I think in openssh's case this is supported by very 
historically very long LoginGraceTime default settings (2 minutes) and a lack
of random early drop usage for MaxStartups.

While you could argue that this is not per-se an openssh security issue, the 
default settings aid here to a trivial denial of service attack against
ssh installations by all linux distributions I've seen.

The result for a user who tries to login is this:
ssh_exchange_identification: Connection closed by remote host

The openssh maintainers actually agree here and it resulted in the following 
changes:
http://www.openbsd.org/cgi-bin/cvsweb/src/usr.bin/ssh/servconf.c?r1=1.234#rev1.234
http://www.openbsd.org/cgi-bin/cvsweb/src/usr.bin/ssh/sshd_config.5?r1=1.156#rev1.156
http://www.openbsd.org/cgi-bin/cvsweb/src/usr.bin/ssh/sshd_config?r1=1.89#rev1.89

I personally don't mind whether this get's a CVE id or not,but considering 
that dropbear got one in the past,I thought I'd bring this up.

Kind regards
Nico
-- 
Nico Golde - http://www.ngolde.de - nion@...ber.ccc.de - GPG: 0xA0A0AAAA

View attachment "sshext.c" of type "text/x-csrc" (3925 bytes)

Content of type "application/pgp-signature" skipped

Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.